hcq/MaxScale
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

MXS-1412: Process response buffers individually

Browse Source 
By processing each buffer individually, the need to iterate over the whole
resultset is removed. Profiling showed that most of the time was spent
navigating the linked list of buffers when an offset into the whole
resultset was used instead of an offset to the individual response buffer.
This commit is contained in:
Markus Mäkelä
2017-09-14 01:00:12 +03:00
parent 49ab5797a2
commit 7e48941a04
1 changed files with 17 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
server/modules/filter/maxrows/maxrows.c
View File

@ -211,7 +211,7 @@ static void maxrows_session_data_free(MAXROWS_SESSION_DATA *data);
static int handle_expecting_fields(MAXROWS_SESSION_DATA *csdata);
static int handle_expecting_nothing(MAXROWS_SESSION_DATA *csdata);
static int handle_expecting_response(MAXROWS_SESSION_DATA *csdata);
static int handle_rows(MAXROWS_SESSION_DATA *csdata);
static int handle_rows(MAXROWS_SESSION_DATA *csdata, GWBUF* buffer, size_t extra_offset);
static int handle_ignoring_response(MAXROWS_SESSION_DATA *csdata);
static bool process_params(char **options,
MXS_CONFIG_PARAMETER *params,
@ -458,7 +458,7 @@ static int clientReply(MXS_FILTER *instance,
break;
case MAXROWS_EXPECTING_ROWS:
rv = handle_rows(csdata);
rv = handle_rows(csdata, data, 0);
break;
case MAXROWS_IGNORING_RESPONSE:
@ -608,7 +608,7 @@ static int handle_expecting_fields(MAXROWS_SESSION_DATA *csdata)
}
csdata->state = MAXROWS_EXPECTING_ROWS;
rv = handle_rows(csdata);
rv = handle_rows(csdata, csdata->res.data,csdata->res.offset);
break;
default: // Field information.
@ -767,28 +767,29 @@ static int handle_expecting_response(MAXROWS_SESSION_DATA *csdata)
*
* @param csdata The maxrows session data.
*/
static int handle_rows(MAXROWS_SESSION_DATA *csdata)
static int handle_rows(MAXROWS_SESSION_DATA *csdata, GWBUF* buffer, size_t extra_offset)
{
ss_dassert(csdata->state == MAXROWS_EXPECTING_ROWS);
ss_dassert(csdata->res.data);
int rv = 1;
bool insufficient = false;
size_t buflen = csdata->res.length;
size_t offset = extra_offset;
size_t buflen = gwbuf_length(buffer);
while (!insufficient && (buflen - csdata->res.offset >= MYSQL_HEADER_LEN))
while (!insufficient && (buflen - offset >= MYSQL_HEADER_LEN))
{
bool pending_large_data = csdata->large_packet;
// header array holds a full EOF packet
uint8_t header[MYSQL_EOF_PACKET_LEN];
gwbuf_copy_data(csdata->res.data,
csdata->res.offset,
gwbuf_copy_data(buffer,
offset,
MYSQL_EOF_PACKET_LEN,
header);
size_t packetlen = MYSQL_HEADER_LEN + MYSQL_GET_PAYLOAD_LEN(header);
if (csdata->res.offset + packetlen <= buflen)
if (offset + packetlen <= buflen)
{
/* Check for large packet packet terminator:
* min is 4 bytes "0x0 0x0 0x0 0xseq_no and
@ -800,10 +801,8 @@ static int handle_rows(MAXROWS_SESSION_DATA *csdata)
packetlen < MYSQL_EOF_PACKET_LEN))
{
// Update offset, number of rows and break
csdata->res.offset += packetlen;
offset += packetlen;
csdata->res.n_rows++;
ss_dassert(csdata->res.offset == buflen);
break;
}
@ -817,9 +816,7 @@ static int handle_rows(MAXROWS_SESSION_DATA *csdata)
// Mark the beginning of a large packet receiving
csdata->large_packet = true;
// Just update offset and break
csdata->res.offset += packetlen;
ss_dassert(csdata->res.offset == buflen);
offset += packetlen;
break;
}
else
@ -834,8 +831,7 @@ static int handle_rows(MAXROWS_SESSION_DATA *csdata)
switch (command)
{
case 0xff: // ERR packet after the rows.
csdata->res.offset += packetlen;
ss_dassert(csdata->res.offset == buflen);
offset += packetlen;
// This is the end of resultset: set big packet var to false
csdata->large_packet = false;
@ -874,8 +870,7 @@ static int handle_rows(MAXROWS_SESSION_DATA *csdata)
* NOTE: not supported right now
*/
case 0xfe: // EOF, the one after the rows.
csdata->res.offset += packetlen;
ss_dassert(csdata->res.offset == buflen);
offset += packetlen;
/* EOF could be the last packet in the transmission:
* check first whether SERVER_MORE_RESULTS_EXIST flag is set.
@ -940,7 +935,7 @@ static int handle_rows(MAXROWS_SESSION_DATA *csdata)
case 0xfb: // NULL
default: // length-encoded-string
csdata->res.offset += packetlen;
offset += packetlen;
// Increase res.n_rows counter while not receiving large packets
if (!csdata->large_packet)
{
@ -972,6 +967,8 @@ static int handle_rows(MAXROWS_SESSION_DATA *csdata)
}
}
csdata->res.offset += offset;
return rv;
}